Workshop paper / 01NeurIPS2026AIWILD & IAB
Before Agents Act: Interpreting Failures under Socially Distributed Access
NeurIPS 2026 Workshop AIWILD & IAB
Upcoming
Independent AI research / Public benefit
San Francisco, CAREDMINDRESEARCH FOUNDATION
Better questions.
More responsible AI.
We study how AI agents reason, remember, and work with people. Open research to make their capabilities and limits easier to understand.
Fig. 01 / Connected context
Conceptual illustration
01 / Publications
THE WORK.
Research should be open to scrutiny.
Start with the papers.
Workshop paper / 02NeurIPS2026TAE & Social Agent
Where Knowledge and Authority Sit Changes What an Agent Benchmark Can Resolve
NeurIPS 2026 Workshop TAE & Social Agent
Workshop paper / 03COLM2026Context Beyond
the Window
the Window
Scoped Verification for Reliable Long-Horizon Agentic Context Evolution under Distribution Shift
COLM 2026 (Context Beyond the Window)
Poster
Workshop paper / 04COLM2026Workshop on
Agent Behavior
Agent Behavior
Evaluating Generative Agents with Actions Grounded in Socially Distributed Task Environments using Incognita
COLM 2026 (Workshop on Agent Behavior)
Poster
Workshop paper / 05ACL2026SURGeLLM
Read paperOpenReview
Selected work on agentic context, verification, and socially grounded evaluation.
02 / Research directions
THE QUESTIONS.
Understanding AI means looking closely
at what happens along the way.
01
Agentic systems
What survives
a long task?
How agents preserve goals and context across multi-step work, use tools, and recover when a plan breaks down.
Context & memory research02
Model evaluation
What does a
score leave out?
Evaluations that make model behavior, failure modes, and actions in socially distributed environments inspectable.
Agent evaluation research03
Human-AI collaboration
Where does human
judgment belong?
Workflows that connect human feedback and AI capabilities while keeping responsibility and accountability with people.
Our operating principles03 / The foundation
PUBLIC
KNOWLEDGE.
PUBLIC GOOD.
Advanced AI should be easier to understand, evaluate, and hold accountable.
RedMind Research Foundation is a California nonprofit organization recognized by the IRS under Section 501(c)(3). We conduct open scientific and educational research on AI systems for public benefit.
Our work is intended for researchers, students, builders, policymakers, and everyone with a stake in how AI is developed and used.
- Based in
- San Francisco, California
- Organization
- Nonprofit research foundation
- Federal status
- IRS-recognized 501(c)(3) public charity
04 / How we work
HUMANS HOLD
THE RESPONSIBILITY.
Human-governed.
Agent-collaborative.
Open-by-default.
01Human authority
Board members and officers retain responsibility for mission, compliance, conflicts, spending, filings, public claims, and formal organizational decisions.
People govern the work.02Bounded agent work
AI agents may research, monitor, summarize, draft, maintain internal trackers, and recommend options. They do not bind or externally represent the Foundation.
Agents support the research.03Open by default
We intend to share research, methods, benchmarks, tools, and educational resources publicly where appropriate, with exceptions for privacy, safety, legal, and confidential matters.
Knowledge serves the public.Our founding operating framework evolves as formal board policies are adopted.
Open research. Shared understanding.
Read our publications05 / Stay connected
GOOD RESEARCH
STARTS WITH
A CONVERSATION.
Ideas, questions, and research conversations are welcome.
Research & general inquiriescontact@
San Francisco, California
Open AI research for public benefit.